Shamik Basu | Data Scientist | MS Data Science USC

Work

Projects & Impact

Production systems and research built from scratch

1st Place — UCLA SAIRS Hackathon 2025 · Sustainability Track

April 2025

Built an end-to-end Streamlit application using Gemini 2.5 Flash to analyze receipts and activity logs, estimating CO₂ emissions per item and generating ranked sustainability recommendations.
Designed the full pipeline: OCR extraction → emission-factor mapping → GenAI recommendations → interactive Plotly dashboard.
Presented live to industry panelists from Microsoft, IBM, NVIDIA, Google, and LinkedIn.

Python Gemini 2.5 Flash Streamlit FastAPI Plotly OCR RAG

Live Demo → GitHub →

Jan – Feb 2025

Engineered custom CUDA kernels for matrix multiplication and image convolution, benchmarking CPU, naïve CUDA, tiled CUDA (shared memory), and cuBLAS across matrix sizes up to N=2048 on NVIDIA Tesla T4 GPUs.
Compiled optimized kernels into .so shared libraries with Python bindings, achieving exponential throughput gains and enabling GPU acceleration inside standard data science workflows.

CUDA C++ cuBLAS Python Jupyter GPU Computing

GitHub →

2024

End-to-end retail analytics pipeline on a 10,000+ order US superstore dataset — EDA, regional trend analysis, customer segmentation, and a 90-day sales forecast.
Built interactive HTML report with Plotly; forecast model achieved MAE ~$1,200 on 90-day holdout.

Python pandas scikit-learn Plotly Matplotlib Jupyter

GitHub → Live Demo →

2022

Real-time parking slot allocation system using shortest-path algorithms on a live 2D map with tkinter GUI and SQLite backend.
Modular 4-layer architecture: GUI, allocation engine, authentication, and map renderer.

Python CUDA Algorithms SQLite tkinter

GitHub →

Experience

Data Science Associate Intern Current

KCC Capital Partners · Los Angeles, CA · Jan 2026 – Present

Fine-tuning and integrating an open-source SLM into a production JavaScript/Docker chatbot service to automate client service request classification and routing, reducing handling overhead for the automation team.
Evaluating model outputs against baseline response quality benchmarks to guide iteration on prompt design and fine-tuning parameters.

Data Scientist

Bajaj Finserv Health · Pune, India · Nov 2023 – Dec 2024

Architected a real-time medical document analytics system using RAG, LangChain, and REST APIs — processing 5M+ records/month at 92% extraction accuracy, replacing a fully manual workflow.
Engineered LLM-based inference pipelines (GPT-3.5 Turbo) to automate high-complexity decision workflows, cutting operational costs by 72%.
Built modular monitoring pipelines with LangChain and Langfuse for model observability, reducing GPU compute utilization by 15%.
Integrated ML model outputs into Power BI dashboards, cutting ad-hoc reporting turnaround time by 42% and enabling self-serve analytics for stakeholders.

Associate Data Scientist

Bajaj Finserv Health · Pune, India · Jul 2022 – Oct 2023

Developed a supervised ML model (Logistic Regression) for workforce performance prediction, improving efficiency outcomes by 22%.
Redesigned the NER-based name-matching algorithm in the fraud detection pipeline, increasing policyholder identification accuracy by 27% and reducing false positives.
Processed 10M+ records in Azure Synapse using SQL to deliver business intelligence reports for senior stakeholders.

Data Engineer Intern

Bajaj Finserv Health · Pune, India · Jan 2022 – Jun 2022

Ran A/B tests and cohort analyses identifying key user behavior patterns that improved web conversion by 37%.
Designed a distributed analytics system in C++, Trino, and Docker supporting 10M+ records across 200+ features.

Data Engineer Intern

Reomnify · Nov 2020 – Jan 2021